[U-Boot] FSL DDR3/4 wrlvl_override question

[York Sun]

On 09/03/2015 11:09 AM, Joakim Tjernlund wrote:
> On Thu, 2015-09-03 at 10:44 -0500, York Sun wrote:
>> Jocke,
>>
>> On 09/03/2015 09:55 AM, Joakim Tjernlund wrote:
>>> in drivers/ddr/fsl/options.c we have:
>>>
>>> #if defined(CONFIG_SYS_FSL_DDR3) || defined(CONFIG_SYS_FSL_DDR4)
>>> 	/*
>>> 	 * due to ddr3 dimm is fly-by topology
>>> 	 * we suggest to enable write leveling to
>>> 	 * meet the tQDSS under different loading.
>>> 	 */
>>> 	popts->wrlvl_en = 1;
>>> 	popts->zq_en = 1;
>>> 	popts->wrlvl_override = 0;
>>> #endif
>>>
>>> Here one disable wrlvl_override no matter what board code wants.
>>> However board code still sets wrlvl_override as it needs specify
>>> a good start value for WRLVL_START DQS[0], just like one do for
>>> the other DQS[X] in WRLVL_CNTL_2/3
>>>
>>> How about remove the above popts->wrlvl_override = 0; line?
>>
>> You don't need to. This is the default to start with. See the call at the end of
>> this function, fsl_ddr_board_options()? It calls for board function to overwrite
>> the default.
> 
> Gahh, missed that one :)
> 
>>
>>>
>>>
>>>
>>> Furthermore drivers/ddr/fsl/ctrl_regs.c I think set_ddr_wrlvl_cntl()
>>> should be adjusted to(or similar):
>>>
>>> --- a/drivers/ddr/fsl/ctrl_regs.c
>>> +++ b/drivers/ddr/fsl/ctrl_regs.c
>>> @@ -2098,7 +2098,7 @@ static void set_ddr_wrlvl_cntl(fsl_ddr_cfg_regs_t *ddr, unsigned int wrlvl_en,
>>>         /* WRLVL_WLR: Write leveling repeition time */
>>>         unsigned int wrlvl_wlr = 0;
>>>         /* WRLVL_START: Write leveling start time */
>>> -       unsigned int wrlvl_start = 0;
>>> +       unsigned int wrlvl_start = 8;
>>>  
>>>         /* suggest enable write leveling for DDR3 due to fly-by topology */
>>>         if (wrlvl_en) {
>>> @@ -2131,10 +2131,11 @@ static void set_ddr_wrlvl_cntl(fsl_ddr_cfg_regs_t *ddr, unsigned int wrlvl_en,
>>>                  * Override the write leveling sample and start time
>>>                  * according to specific board
>>>                  */
>>> -               if (popts->wrlvl_override) {
>>> -                       wrlvl_smpl = popts->wrlvl_sample;
>>> -                       wrlvl_start = popts->wrlvl_start;
>>> -               }
>>> +       }
>>> +
>>> +       if (popts->wrlvl_override) {
>>> +               wrlvl_smpl = popts->wrlvl_sample;
>>> +               wrlvl_start = popts->wrlvl_start;
>>>         }
>>>
>>>
>>> so that board not using auto WRLVL a chance to specify these parameters.
>>>
>>> I am no expert on DDR so I might be totally off here ...
>>>
>>
>> NACK. Your change is wrong. wrlvl_smpl and wrlvl_start only have meaning if
> 
> OK, I was unsure about wrlvl_start as it was not spelled out in RM like it is for the
> other bits in this register.
> 
>> wrlvl_en is set. You can make DDR work without write leveling by setting a lot
>> of timing registers. It is not easy and requires extensive experience and
>> exercise. It is not recommended for most users. If you have to use that way,
>> please work with an FAE to get all the correct settings.
> 
> We have this coldstart problem that often DDR init fails, we get "Waiting for D_INIT timeout.
> Memory may not work." and then the board is stuck.
> 
> Working with FAE trying to find the cause it was suggested we try specify the values
> ourself (using CW and a probe to auto calculate these). So far this has not worked at
> all so I think we will stop here with that track.
> 
> Any idea where to look? So far all kinds of HW measurements has been done, all looks ok.
> Once one gets past this fail(new coldstart or warmstat) RAM works with anything we
> can throw at it, even rebuilt @world (gentoo) on the board.
>   

First thing to check is the DDR reset signal, the one resets the DDR DIMM.
Second, you mentioned CW tool, does that pass?
Third, you can hard-code the registers you got from CW into u-boot, totally
bypass this driver.

York